Loading...
Welcome to Journal of University of Chinese Academy of Sciences,Today is

Current Issue

2024, Vol.41, No.5 Previous Issue    Next Issue
Innovation Article
An adaptive variance reduction method with negative momentum
LIU Hai, GUO Tiande, HAN Congying
2024, 41 (5): 577-588.  DOI: 10.7523/j.ucas.2024.024
Abstract ( 522 ) PDF (0KB) ( 0 )
Stochastic variance reduction methods have been successful in solving large scale machine learning problems, and researchers cooperate them with adaptive stepsize schemes to further alleviate the burden of parameter-tuning. In this article, we propose that there exists a trade-off between progress and effectiveness of adaptive stepsize arising in the SVRG-BB algorithm. To enhance the practical performance of SVRG-BB, we introduce the Katyusha momentum to handle the aforementioned trade-off. The linear convergence rate of the resulting SVRG-BB-Katyusha algorithm is proven under strong convexity condition. Moreover, we propose SVRG-BB-Katyusha-SPARSE algorithm which uses Katyusha momentum sparsely in the inner iterations. Numerical experiments are given to illustrate that the proposed algorithms have promising advantages over SVRG-BB, in the sense that the optimality gaps of the proposed algorithms are smaller than the optimality gap of SVRG-BB by orders of magnitude.
Research Articles
Dynamic evolution of natural convection in a porous square cavity
ZHANG Lianping, WANG Shimin
2024, 41 (5): 589-603.  DOI: 10.7523/j.ucas.2022.077
Abstract ( 284 ) PDF (0KB) ( 0 )
Natural convection in fluid-saturated porous cavities is a classical problem in the study of nonlinear hydrodynamics. Limited by the early computing capabilities, previous studies mainly focused on the dynamic mechanism of the spatiotemporal evolution of centrosymmetric convection modes, and could not fully describe the dynamic behavior of real-world porous natural convection systems with all convection modes. Therefore, the available results could not be directly applied to practical engineering problems. In this study, two-dimensional natural convection in a porous square cavity heated from below is numerically simulated with complete convection modes based on the Galerkin method and the collocation pseudo-spectral method, and the complete route for natural convection dynamically evolving from onset towards chaos is obtained for the first time. The modeling results reveal that along with Ra increasing from 4π2 to 1 200, the natural convection evolves across 21 different stages (including one steady convection stage, six single-frequency oscillation stages, nine quasiperiodic oscillation stages, two frequency-locked resonance stages, and three aperiodic chaotic oscillation stages). The spatial variation and temporal oscillation of convection modes are the fundamental reasons for the repeated alternation between different convection patterns. Different characteristics of single-frequency, quasiperiodic, frequency-locked, and chaotic oscillations in the calculated heat flow represented by Nu are systematically compared in terms of four graphical methods, viz. time series, power spectrum, phase portrait, and Lorenz mapping. The calculated average heat flow and primary oscillating frequency as functions of Ra can be fitted by simple analytical formulas, approximately obeying the scaling laws given by the classical boundary layer theory in a wide range of Ra values. This provides convenient calculation tools for practical engineering designs and applications under the corresponding conditions.
Dynamic friction polishing of single crystal diamond
ZHANG Haochen, XU Kai, YAN Zengyu, SONG Zhipeng, CHEN Guangchao
2024, 41 (5): 604-611.  DOI: 10.7523/j.ucas.2023.003
Abstract ( 306 ) PDF (0KB) ( 0 )
The dynamic friction polishing (DFP) method has received extensive attention due to its significantly high removal rate comparing with other single crystal diamond polishing methods. In this study, three parameters (polishing load, polishing time, and polishing plate linear velocity) and three clamping types (bonded, inlay, and caliper type) were studied on the influence on the samples’ surface roughness (Ra) improvement and the mass loss. The results showed that the increase of the mass loss and the reduction of roughness occurred with increasing the polishing load, the polishing time, and the polishing plate linear velocity, respectively. Whilst, the X-ray diffraction intensity of (400) crystalline surface of each polished sample was enhanced. Among the three clamping types, the bonded clamping type possessed the fastest removal rates, up to 25.2 nm/h. A polishing evaluation parameter, K, was defined as KRam, i.e., representing the improvement in surface roughness obtained per unit mass loss. According to variation of the parameter K, the DFP procedure could be divided into two stages, “the roughness improvement domination stage” and “the mass loss domination stage”, with the former having higher K values than the latter. K values increased monotonically with the increase of the polishing plate linear velocity, whilst they were influenced complicatedly by the polishing load and the polishing time.
Influence and driving force of administrative division adjustment on urban spatial expansion: a case study of withdrawal from city into district in Jiutai, Changchun City
DONG Yaojia, WANG Fuyuan, WANG Kaiyong
2024, 41 (5): 612-624.  DOI: 10.7523/j.ucas.2022.080
Abstract ( 399 ) PDF (0KB) ( 0 )
Removing counties (cities) into districts is one of the main modes of administrative division adjustment, which has an impact on the urban spatial structure. Using multi-temporal remote sensing images of Changchun, administrative division vector graphics and statistical survey data of social and economic conditions, this paper analyzes the evolution of construction land and economic development in Changchun City before and after Jiutai was withdrawn from the city and divided into districts. Utilizing ENVI software, GIS technology, and other means, this paper revealed the impact of Jiutai’s withdrawal from the city and the establishment as a district on the spatial expansion of Changchun City at the two-level spatial scale of Changchun City and Jiutai District. The research found that: 1) Dismantling the city into districts accelerated the expansion process of Jiutai District, which further affected the direction and extent of urban spatial expansion in Changchun City. 2) The decommissioning of the city into districts caused Jiutai and the main urban area of Changchun to converge towards each other, and the spatial distance between the two districts gradually narrowed. 3) The “polarization effect” of the Jiutai District became more pronounced after the city was withdrawn and divided into districts, and high-density clusters appeared in the main urban area of Changchun. 4) The division of counties (cities) into districts is an important driving force for urban spatial expansion, primarily reflected in three aspects: policy, industry, and land use.
Remote sensing classification and extraction of alpine wetlands information in summer in Dam Qu Watershed, the source of Yangtze River
XIE Wanrong, LIU Shaochuang, WU Yunjia, ZHANG Shuo
2024, 41 (5): 625-635.  DOI: 10.7523/j.ucas.2023.031
Abstract ( 229 ) PDF (0KB) ( 0 )
As the real source of the Yangtze River, Dam Qu Watershed is covered with a large number of alpine wetlands. Due to the dispute on the determination of the source of the Yangtze River in early years and difficulties in scientific research caused by the harsh natural environment, Dam Qu Watershed has not received widespread attention.Consequently, research on wetlands in this area is limited, particularly in high-resolution wetland classification, leaving a significant gap in the literature. Based on Sentinel-2B images from August 19, 2020, the information on each category of alpine wetlands in Dam Qu Watershed was extracted by using the hierarchical classification method of object-oriented features, and the horizontal and vertical distribution characteristics of wetlands were summarized in combination with the surface morphology and texture of the study area. The results show that the alpine wetlands in Dam Qu Watershed are divided into three first-level categories and nine second-level categories, with a total area of 3 364.74 km2, of which the main part is marsh wetlands, with an area of 2 908.94 km2, accounting for 86.45% of the total area of wetlands. The areas of river wetlands and lake wetlands are similar, at 269.28 km2 and 186.52 km2, respectively. In addition to alpine wetlands, there are permanent glaciers and snow in the watershed, with an area of 107.17 km2.The horizontal distribution of wetlands in the watershed is generally characterized by more wetlands in the southeast and fewer in the northwest. These wetlands are mainly concentrated in the flat beach surrounded by hills or the catchment area with low terrain and poor drainage, as well as the foot of shady slope. In the vertical direction, alpine wetlands are mainly distributed in the range of 4 650-5 100 m above sea level. As the terrain rises, wetland resources gradually decrease. Areas above 5 400 m are mainly covered by glaciers and snow, with almost no alpine wetland. This study provides a scientific basis for the effective protection and rational utilization of wetland resources in the source of the Yangtze River.
Radiometric correction of multispectral camera based on filter wheel
LIU Yang, FANG Junyong, LIU Xue, WANG Xiao, ZHANG Xiaohong
2024, 41 (5): 636-643.  DOI: 10.7523/j.ucas.2022.083
Abstract ( 281 ) PDF (0KB) ( 0 )
Aiming at the problems of limited acquisition method, band registration, and data quantification of near-surface multispectral images, this paper introduces a self-developed filter wheel multispectral camera that can work continuously in the field, has simple operation and easy band registration, and completes relative radiometric correction and absolute radiometric calibration with high-precision integrating sphere in the laboratory. The multi-stage brightness calibration method and empirical linear method are used to establish a calibration model to correct the uneven response of multi-band multispectral camera and calculate the spectral radiation brightness at the pupil. In this paper, we design data acquisition and processing experiments of different brightness and integration time, and give the optimal integration time range for each band of the new camera. The results of radiometric correction and calibration show that the selected correction coefficient has high stability and reliability for multispectral images under different imaging conditions, and is suitable for radiometric correction of filter wheel-type multispectral cameras. The calibration results can be used to quantitatively retrieve the target parameters of the land surface, which has broad application prospects in the fields of crop monitoring, environmental monitoring and protection.
Chirp scaling algorithm based on fractional Fourier transform and image weighted entropy
SHANG Min, XU Xianghui
2024, 41 (5): 644-653.  DOI: 10.7523/j.ucas.2022.084
Abstract ( 312 ) PDF (0KB) ( 0 )
In order to solve the problem of Doppler parameters varying with skew and low image resolution in the traditional chirp scaling (CS) imaging algorithm based on Fourier transform and matched filtering, an algorithm to optimize CS imaging algorithm using fractional Fourier transform (FRFT) is proposed. Firstly, the echo signal model of squint synthetic aperture radar (SAR) is established, and the echo signal model is derived using FRFT instead of matched filtering. To search for the optimal azimuth rotation angle, the cost function of the image is established according to the weighted minimum entropy, and the gradient descent optimization algorithm of the momentum method is used for iterative calculation. Finally, a higher-resolution SAR image is obtained. To verify the effectiveness of the algorithm, experiments were carried out on point target simulation data and measured SAR data sets respectively. The results show that, compared with the traditional CS imaging algorithm, the proposed algorithm achieves a narrower main lobe width, lower sidelobe, and clearer images.
Building extraction method based on MFF-Deeplabv3+ network for high-resolution remote sensing images
CHEN Jingwei, LI Yu, CHEN Jun, ZHANG Hongqun
2024, 41 (5): 654-664.  DOI: 10.7523/j.ucas.2023.010
Abstract ( 516 ) PDF (0KB) ( 0 )
Automatic extraction of building information from high-resolution remote sensing images is of great significance in the fields of environmental monitoring, earthquake mitigation, and land use, making it a research hotspot in the field of high-resolution remote sensing applications. In order to improve the accuracy of building extraction from high-resolution remote sensing images, a building extraction method based on MFF-Deeplabv3+(multiscale feature fusion-Deeplabv3+) network for high-resolution remote sensing images is proposed in this paper. First, the multi-scale feature enhancement module is designed to enable the network to capture more scale context information; then, the feature fusion module is designed to effectively fuse deep features with shallow features to reduce the loss of detail information; finally, the attention mechanism module is introduced to select accurate features adaptively. In the comparison experiments of the Inria building dataset, MFF-Deeplabv3+ achieved the highest accuracy in PA, MPA, FWIoU, and MIoU metrics with 95.75%, 91.22%, 92.12%, and 85.01%, respectively, while the generalization experiments of the WHU building dataset achieved good results. The results show that this method extracts building information from high-resolution remote sensing images with high accuracy and strong generalization.
A trajectory planning based on characteristic parameter fitting for multi-ground targets of the ground tracks of two satellites
HUANG Xiongwei, WANG Shuquan, ZHANG Yang, HE Shengmao
2024, 41 (5): 665-676.  DOI: 10.7523/j.ucas.2022.070
Abstract ( 335 ) PDF (0KB) ( 0 )
In order to solve the problems related to the ground tracks of satellites for Earth observation with multiple orbital maneuvering capabilities, multi-detection targets, and multi-detector coordination, a fast calculation method of the ground tracks of satellites based on characteristic parameter fitting is proposed. An optimization model is established based on the database about the splicing method of the ground tracks of satellites. Firstly, the characteristic parameters describing the ground tracks of satellites are analyzed and refined, and polynomials are used to fit these characteristic parameters to achieve fast calculation. Then, an observation code database is established according to the detection target, and the database is further organized into a trajectory database. Finally, an optimization model based on the trajectory database is established to solve all problems on the ground tracks of satellites. The simulation of a fast full coverage task for a fixed target in CTOC11 (the 11th China Trajectory Optimization Competition) was completed within 19.19 days. The results show that the design method and scheme adopted in this paper can be used for the design of ground tracks of satellites in various near-circular orbits.
A pose estimation algorithm for spatial non-cooperative targets based on point cloud registration
GUO Sujie, GUO Chongbin
2024, 41 (5): 677-686.  DOI: 10.7523/j.ucas.2023.027
Abstract ( 374 ) PDF (0KB) ( 0 )
Aiming at pose estimation for non-cooperative targets in on-orbit maintenance operations of space robots, we propose a pose estimation algorithm based on point cloud registration. Firstly, the hybrid filtering algorithm preserves the shape of the point cloud to the greatest extent while reducing its density. Then, the principal component analysis algorithm is used to establish the eigenvector transformation. The RANSAC algorithm is employed for coarse registration, followed by the improved ICP algorithm for fine registration, which results in the estimation of the rotation matrix, translation matrix, and attitude. Experiments are carried out to evaluate the performance of the algorithm. Simulated satellite point cloud models processed by rotation-translation transformations and Gaussian noise are used to verify the point cloud registration performance. The satellite model scene collected by a TOF camera point cloud is used to validate the pose estimation algorithm. The results show that the proposed algorithm has improved anti-noise performance, and showed higher robustness compared to traditional registration algorithms, with rotation attitude angle error less than 0.4° and displacement error less than 3 mm.
Point cloud compression of deep learning based on multi-scale feature and attention mechanism
HUANG Yulin, LIANG Lei, LI Weijun, XI Xiaohuan
2024, 41 (5): 687-694.  DOI: 10.7523/j.ucas.2023.077
Abstract ( 318 ) PDF (0KB) ( 0 )
3D point clouds have extensive applications in the auto-drive, 3D real scene, and other fields. But complex scene requires massive point clouds to represent which brings great challenges to storage space, data processing and transmission bandwidth. A multi-scale attention point cloud geometry compression (MSA-GPCC) is proposed to compress point cloud data based on multi-scale features, attention mechanism, and variational auto-encoders (VAE). Experiments and analysis are carried out based on MPEG data sets. The results show that MSA-GPCC performs better than those of the traditional G-PCC and deep-learning-based D-PCC algorithms, D1 BD-PSNR is improved by 7.72 and 4.91 dB respectively, and D2 BD-PSNR is improved by 5.56 and 3.09 dB respectively.
Central node selection strategy of spatial robot cluster based on data and compression ratio prediction
YANG Xuan, CHEN Hongyu
2024, 41 (5): 695-704.  DOI: 10.7523/j.ucas.2023.007
Abstract ( 149 ) PDF (0KB) ( 0 )
In recent years, the in-orbit service technology of space robot clusters has attracted the attention of various space powers. When the space robot cluster serves the target spacecraft in orbit, the collected target information needs to be transmitted to the central satellite. How to balance the communication power consumption of each node in the cluster is an important research problem. Aiming at the problem of optimal communication power between the space robot cluster and the data hub satellite, a central node selection algorithm (data and data compression ratio prediction,DCP) based on data and compression ratio prediction was proposed in this paper. Since the communication power consumption in the cluster communication is mainly related to the communication distance and communication duration (data volume), the data and compression rate at future times can be predicted based on the movement trajectory of the cluster, thus selecting the optimal central node of the cluster and constructing the communication link. In the experimental simulation, compared to a fixed point, degree centrality, betweenness centrality, and closeness centrality, DCP algorithm can effectively reduce the power consumption of cluster communication, and the error is less than 3% compared with the actual optimal result.
Seamless image completion via GAN inversion
YU Yongsheng, LUO Tiejian
2024, 41 (5): 705-714.  DOI: 10.7523/j.ucas.2022.075
Abstract ( 570 ) PDF (0KB) ( 0 )
Image completion is widely used in unwanted object removal and media editing, which aims to find a semantically consistent way to recover corrupted images. This paper is based on generative adversarial network (GAN) inversion, which leverages a pre-trained GAN model as an effective prior to filling in the missing regions with photo-realistic textures. However, existing GAN inversion methods ignore that image completion is a generative task with hard constraints, making final images have noticeable color and semantic discontinuity issues. This paper designs a novel bi-directional perceptual generator and pre-modulation network to seamlessly fill in the images. The bi-directional perceptual generator uses extended latent space to help the model perceive the non-missing regions of the input images in terms of data representations. The pre-modulated networks utilize a multiscale structure further providing more discriminative semantics for the style vectors. In this paper, experiments are conducted on Places2 and CelebA-HQ datasets to verify that the proposed method builds a bridge between GAN inversion and image completion and outperforms current mainstream algorithms, especially in FID metrics up to 49.2% enhancement at most.
Brief Report
Low-complexity CA-SCL decoding optimization algorithm for low-orbit satellite communication
HU Xiuqi, HOU Huiling, LIANG Guang, YU Jinpei
2024, 41 (5): 715-720.  DOI: 10.7523/j.ucas.2023.016
Abstract ( 340 ) PDF (0KB) ( 0 )
Based on the low-orbit satellite communication environment with limited hardware resources, this paper uses polarization code as the channel coding method, and combines the successive redundancy check (CRC) assisted successive cancellation list (SCL) decoder (CA-SCL) with critical sets and adaptive algorithms to propose an optimized CA-SCL decoder (OCASCL). The performance of the OCASCL decoder is better than the classic SCL decoder, the computational complexity can be reduced by 65%-70%.